Plan and manage an Azure AI solution 1.1 Choose the appropriate Foundry services for generative AI and agents
A product team wants its first customer-service agent running quickly in Microsoft…
Study guide skill: Choose the appropriate Foundry services for generative tasks, grounding, vector search, agent workflows, or multimodal processing
Question B1-001
A product team wants its first customer-service agent running quickly in Microsoft Foundry. The agent only needs instructions, a model, and a few built-in tools, and the team doesn't want to write, package, or operate any agent code. Which option should you choose?
Select one answer.
Answer: A
Prompt agents are defined entirely through configuration (instructions, model and tools) and Foundry runs them with no code or infrastructure to manage. Hosted agents are for bringing your own code as a container, and a custom AKS app or a fine-tuned model adds code and operations the team wants to avoid.
Deep dive
How it works
Foundry Agent Service has three agent types, and your choice decides how much you build and how much Foundry manages. A prompt agent is defined entirely through configuration: instructions, a model and tools. Foundry runs it for you. There's no application code to maintain, and no containers or packages to optimize, scale or monitor. You can create one in the Foundry portal, or define it with the SDKs or REST API so it fits a CI/CD pipeline.
The docs say prompt agents are best for getting started fast, internal tools, production agents that don't need custom orchestration logic, and teams that want a managed runtime without infrastructure overhead. Tools such as web search, file search, code interpreter, MCP servers and custom functions can be curated once in a toolbox and shared across agents.
If the agent needs spoken conversations, a voice-based prompt agent is the managed option. You configure the model, instructions, audio and tools, and connect to the agent through Voice Live over a real-time WebSocket.
The other options, and when you'd use them
- A fine-tuned model with a custom system message: Fine-tuning adapts a pretrained model through additional training on task-specific data. It adds training and hosting costs, and it doesn't replace retrieval for current information or application-level safety controls. Use it when you have high-quality, task-specific training data and want, for example, a consistent tone, format or schema. It gives you a model, not an agent runtime.
- A custom LangGraph app on AKS that calls the Responses API: Calling the Responses API from your own code builds an ephemeral agent, whose definition lives in your application code. That fits when you already run agent code elsewhere, but you still run and operate that code and its hosting yourself.
- A hosted agent packaged as a container: Hosted agents are code-based agents built with Agent Framework, LangGraph, the OpenAI Agents SDK or your own code. Foundry runs them with a managed endpoint, automatic scaling and a dedicated Entra identity. They're best for agents that call into your own custom code, but you maintain the agent logic.
Exam tip
"No code, just instructions, a model and tools" means a prompt agent. "Bring our own framework code" means a hosted agent, and "agent code already runs elsewhere" means the Responses API.
References
Last reviewed: